2008/09/01 - epics channel access in depth
TRANSCRIPT
www.cosylab.com
Channel Access in Depth
Rok Sabjan09/01/2008
Based on presentation by K. Evans, APS
2
Topics Covered
• Channel Access
- The means by which EPICS Clients and Servers talk to each other
• CaSnooper
- An application to monitor Search Requests for Process variables
• CASW
- An application to monitor Beacon Anomalies
• ParseCASW
- An application to convert CASW output to identifiable events
• RunCaSnooper
- An application that provides an interface to CaSnooper and CASW with an associated MEDM and StripTool
3
EPICS Overview
MEDM MEDM Client Client Client MEDM
Server IOC IOC
Meter Power Supply Camera
IOC
Channel Access
4
Channel Access Concepts
• Network Protocols
• Process Variable Connection Process
• Search Request
• Exist Test
• Beacons
• Beacon Anomaly
• CaRepeater
5
Network Protocols
• Channel Access uses two Network Protocols, UDP and TCP
• UDP (User Datagram Protocol)
- One way, unreliable
- Send out packets, no guarantee they reach their destination
- Can be broadcast or directed (unicasts) Broadcasts: To all IP addresses, e.g. 123.45.6.255 Unicasts: To a specific IP address, e.g. 123.45.6.100
- Broadcasts may not leave subnets for security reasons
• TCP (Transmission Control Protocol)
- Two way, reliable, persistent
- Socket at each end
- Acknowledgements, timeouts, retransmissions, etc. guarantee reliability
6
Connection Process
• A client (e.g. MEDM) wanting a PV sends a UDP search request- Sent to EPICS_CA_ADDR_LIST- (Or its default -- broadcast to all interfaces on the host machine)- Sent on EPICS_CA_SERVER_PORT [5064]- Do you have this PV?
• Each Server that gets a packet does an exist test- Do I have this PV?
• Server with the PV sends a directed UDP reply to the Client - I have this PV.
• A TCP connection is established between the Server and the Client (or an existing one is used)- One per Client-Server pair, no matter how many PVs- Referred to as a Circuit- Let’s talk.
7
Search and Connect Graphically
MEDM MEDM Client Client Client MEDM
Server IOC IOC
Meter Power Supply Camera
IOC
3. TCP Connection
Let’s talk !
1. UDP Broadcast Sequence
Who has it ?
Check Check CheckCheck
2. UDP Reply
I have it !
IOC
8
Search Request
• A client makes a search request when it wants to find out what server has the PV- Happens when a PV is first created in the client- On a beacon anomaly (unresolved PVs only)- When another PV is created (unresolved PVs only)
• A search request consists of a sequence of UDP packets- Starts with a small interval (30 ms), that doubles each time- Until it gets larger than 5 s, then it stays at 5 s
- Stops after 100 packets or when it gets a response- Used to never try again until it sees a beacon anomaly or
creates a new PV As of 3.14.7 retries at a slow rate
- Total time is about 8 minutes to do all 100- The sequence may be different owing to fine tuning
• Usually connects on the first packet or the first few
9
Exist Test
• Every time a Server receives a search request packet, its pvExistTest routine is called
• The Server has to check if it has the PV
- Returns ExistsHere or DoesNotExistHere
• Normally a search request sequence ends after a few packets
- Because one Server soon returns ExistsHere
• For PVs that do not exist
- There are 100 tests per search request sequence for that PV
- This happens every time a Client initiates a search request sequence Each time the Client searches for a new PV At each beacon anomaly, perceived or real
10
• A Beacon is a UDP broadcast packet sent by a Server
• When it is healthy, each Server broadcasts a UDP beacon at regular intervals (like a heartbeat)- EPICS_CA_BEACON_PERIOD, 15 s by default
• When it is coming up, each Server broadcasts a startup sequence of UDP beacons
- Starts with a small interval (25 ms, 75 ms for VxWorks)- Interval doubles each time- Until it gets larger than 15 s, then it stays at 15 s
- Takes about 10 beacons and 40 s to get to steady state
• Clients monitor the beacons- Determine connection status, whether to reissue searches
Beacons
11
Beacon Anomaly
• A Beacon Anomaly is any change from the normal beacon interval (15 s)
• No beacons:- After 30 sec the client sends message over TCP connection- If no beacons and no reply, connection is down- That is when MEDM screens go white
• Abnormal interval:- Short: IOC has come up- Long: IOC was disconnected
• May cause clients to reissue outstanding search requests• Network problems can look like beacon anomalies
12
Virtual Circuit Disconnect
• 3.13 and early 3.14- Hang-up message or no response from server for 30 sec.- If not a hang-up, then client sends “Are you there” query- If no response for 5 sec, TCP connection is closed- MEDM screens go white- Clients reissue search requests
• 3.14.5 and later- Hang-up message from server- TCP connection is closed- MEDM screens go white- Clients reissue search requests
13
Virtual Circuit Unresponsive
• 3.14.5 and later- No response from server for 30 sec.- Client then sends “Are you there” query- If no response for 5 sec, TCP connection is not closed
For several hours, at least- MEDM screens go white- Clients do not reissue search requests
Helps with network storms
- Clients that do not call ca_poll frequently get a virtual circuit disconnect even though the server may be OK Clients written for 3.13 but using 3.14 may have a problem May be changed in future versions
14
CaRepeater
• UDP broadcasts are not guaranteed to go to every process on a workstation
• CaRepeater solves this problem- There is one CaRepeater process per workstation- Clients make a TCP connection to it when they start up- CaRepeater receives the beacons
EPICS_CA_REPEATER_PORT [usually 5065]- CaRepeater forwards the beacons to the Client
• This problem does not exist on most modern systems
Workstation
CaRepeater
MEDM
MEDM
MEDM
15
Multiple Servers on the Same Host
• Used to not be possible at all (Base 3.13)
• Now, it can be done, but there are problems
• Will get message- cas warning: Configured TCP port was unavailable. Using
dynamically assigned TCP port 45003, but now two or more servers share the same UDP port. Depending on your IP kernel this server may not be reachable with UDP unicast (a host's IP in EPICS_CA_ADDR_LIST)
• First part means clients will establish their circuit on another port than the default 5064.
- Not a problem
• Second part means unicast search requests may not get to both servers- UDP deficiency, similar to the CaRepeater problem- May be a problem when EPICS_CA_ADDR_LIST is used
16
Important Environment Variables
• EPICS_CA_ADDR_LIST- Determines where to search- Is a list (separated by spaces)
“123.45.1.255 123.45.2.14 123.45.2.108”- Default is broadcast addresses of all interfaces on the host
Works when servers are on same subnet as clients- Broadcast address
Goes to all servers on a subnet Example: 123.45.1.255 Use ifconfig –a on UNIX to find it (or ask an administrator)
• EPICS_CA_AUTO_ADDR_LIST- YES: Include default addresses above in searches- NO: Do not search on default addresses- If you set EPICS_CA_ADDR_LIST, usually set this to NO
17
EPICS_CA_ADDR_LIST
MEDM MEDM Client Client Client MEDM
Server IOC IOC
Meter Power Supply Camera
IOC
Subnet 2Subnet 1
Specific
123.45.2.108
Broadcast
123.45.1.255
Not Included
18
Other Environment Variables
• CA ClientEPICS_CA_ADDR_LIST
EPICS_CA_AUTO_ADDR_LIST
EPICS_CA_CONN_TMO
EPICS_CA_BEACON_PERIOD
EPICS_CA_REPEATER_PORT
EPICS_CA_SERVER_PORT
EPICS_CA_MAX_ARRAY_BYTES
EPICS_TS_MIN_WEST
• See the Channel Access Reference Manual for more information
• CA ServerEPICS_CAS_SERVER_PORT
EPICS_CAS_AUTO_BEACON_ADDR_LIST
EPICS_CAS_BEACON_ADDR_LIST
EPICS_CAS_BEACON_PERIOD
EPICS_CAS_BEACON_PORT
EPICS_CAS_INTF_ADDR_LIST
EPICS_CAS_IGNORE_ADDR_LIST
19
Summary
• Clients send search requests when they want a PV• Each server has to check if it has the PV for every packet in the
search-request sequence• Servers send beacons at regular intervals and with a faster
pattern when they come up• A beacon anomaly is any pattern that is not a regular beacon• Beacon anomalies may cause clients to resend search
requests for any unresolved PVs• Search request sequences end early for found PVs but not for
non-existent PVs• Search requests put a load on the servers and add to network
traffic- This can cause problems- Consequently, undesirable beacon anomalies and search
requests should be minimized or eliminated• Searches are on port 5064 and beacons are on port 5065
20
CaSnooper
• CaSnooper is a server whose ExistTest routine keeps track of search requests rather than seeing if it has the PV
• It can print the names of all PVs being searched for and related statistics using several report formats
• It can also check if these PVs are connected (C) or not (NC)• It has internal PVs if started with the –n option
- ExistTest rates that can be monitored- Others that allow it to be controlled from an MEDM screen- The PV prefix [default CaSnoop] can be changed to prevent collisions
• Running CaSnooper:- Run at the command line to get one report- Run with PVs for monitoring, say with SDDSmonitor or StripTool- Run with PVs and control with MEDM for continuous operation
• To run CaSnooper you may need the full path- /usr/local/epics/extensions/bin/solaris-sparc/caSnooper- It is not installed at the APS for 3.13 (3.13 servers have less capability)
21
Sample CaSnooper Output
Not connected, will be C for connected (hardly
ever the case)
Print top 10 (-p10)
Check top 10 (-c10)
machine:port, (can be used to identify source)
name
search rate in Hz
statistics
Two lines from RunCaSnooper
individual name, prefix
22
Control CaSnooper via MEDM
Shell command to start CaSnooper, CASW, StripTool, etc.
Use these to set what will happen when you press Report. Case illustrated will print the top 10.
Cartesian plot of requestRate and individualRate
Individual rate for CaSnoop.test, which doesn’t exist
Request rate
Execute selected reports in the CaSnooper stdout
Reset the counters in CaSnooper
Stop CaSnooper
CaSnooper was started here (with EPICS_CA_REPEATER_PORT = 5065)
23
CaSnooper Options
24
CASW
• CASW (Channel Access Server Watcher) monitors Beacon Anomalies
• Is a simple command-line utility• Part of EPICS Base• May need a full path to the
version of base desired- 3.14
(Recommended) /usr/local/epics/base3.14.3/bin/solaris-sparc/casw
- 3.13: /usr/local/epics/base/bin/solaris/casw
• Prints a line with a timestamp when it sees a beacon anomaly
CaSnooper Starting
25
ParseCASW
• CASW produces a list of beacons that came at the wrong time
- Listed in the order they happen Intervals between anomalies are important Not easy to see from the output
- Anomalies from different causes are mixed together
- Cause of an anomaly sequence is not readily apparent
• ParseCASW parses the CASW output
- Uses artificial intelligence to try to determine the event that caused the anomaly (e.g. an IOC coming up)
- Prints the events instead of the anomalies
• Can be used in two ways
- Parse output saved from CASW (or OAG data logging)
- Pipe CASW into ParseCASW in real time
26
ParseCASW Output
Flakey IOC
Most common
Flakey IOC
Long File
Short File
27
ParseCASW Verbose Output
• With the –v option you get more details
- You can decide for yourself what kind of event it is
28
ParseCASW Options
29
StripTool
Search for individual PV ended after about 8 minutes
Reverse Gateways (1 min update)
(Reverse Gateways, CaSnooper, and all IOCs see the same Search Requests)
CaSnooper (1 sec update)
CaSnooper was started here
30
OAG Monitoring
• The search request rates from the Reverse Gateways on the machine subnet are being continuously monitored
• You can can access the history from
http://www.aps.anl.gov/asd/oag/logging/MonitorDataReview.html
31
RunCaSnooper
• RunCaSnooper is a shell script that provides an interface to CaSnooper and CASW with an associated MEDM and StripTool
• Is APS specific, but the script can be modified for your situation- Is part of IocappsTools (Type iocHelp for a list)
• By default it brings up an MEDM and a StripTool and does not start CaSnooper- If CaSnooper is running, the MEDM screen will not be white- If it is white, you can start CaSnooper from the MEDM screen- Uses EPICS_CA_REPEATER_PORT=9876 by default, not 5065- You can also start CASW and StripTool from the MEDM screen
• Everything is generated on the fly and stored in /tmp• Look in /tmp for:
- Logs of the CaSnooper reports and CASW output- MEDM ADL file and StripTool configuration files
• Start it (type runCaSnooper) with no options:- Displays extensive directions - Then optionally allows you to start it (type y to continue)
32
RunCaSnooper
33
RunCaSnooper Options
34
References
• Jeffery O. Hill, EPICS R3.14 Channel Access Reference Manual, (EPICS Documentation, 2005 or latest).
http://www.aps.anl.gov/epics/modules/base/R3-14/index.php
• W. Richard Stevens, UNIX Network Programming, (Prentice-Hall, Upper Saddle River, NJ, 1998) Vol. 1.
• K. Evans, Jr., CaSnooper Reference Manual, (EPICS Documentation, 2005 or latest).
http://www.aps.anl.gov/epics/extensions/caSnooper/index.php